Evaluation Report on the minimum required training labels - post hoc

[raymond301]

Idea generated by Dr. Chen Wang.

Wang, Chen, Ph.D.
1/9/2025 1:32 PM
everyone was asking what's the min N. of cells
technically that's a number in the config file, I set the default to 20. but I suspect your asking about the theoretical minimum. which according to Caitlin's assessment would be 5000. To which there has only ever been 1 PI who annotates to that high of a level.
yes, I thought some retrospective downsample... by Wang, Chen, Ph.D.
Wang, Chen, Ph.D.
1:35 PM
yes, I thought some retrospective downsampling plots of less annotated cells v. performance would help
certainly that'll depend on cell-type by Wang, Chen, Ph.D.
Wang, Chen, Ph.D.
1:35 PM
certainly that'll depend on cell-type
interesting, like another evaluation module... by Moore, Raymond M.
1:36 PM
Moore, Raymond M.
interesting, like another evaluation module directly in the pipeline? That's pretty easy to do.
yes, like after the final model, another sc... by Wang, Chen, Ph.D.
Wang, Chen, Ph.D.
1:37 PM
yes, like after the final model, another script
to do post-hoc retro plots of N. of annotate... by Wang, Chen, Ph.D.
Wang, Chen, Ph.D.
1:37 PM
to do post-hoc retro plots of N. of annotated cells v. perf
because this will inform the next steps, sh... by Wang, Chen, Ph.D.
Wang, Chen, Ph.D.
1:37 PM
because this will inform the next steps, should PIs annotate more
or good enough by Wang, Chen, Ph.D.
Wang, Chen, Ph.D.
1:37 PM
or good enough
has context menu
Wang, Chen, Ph.D. is typing

post-hoc in the way that you can fix the best selected features (assuming feature selection is already done), but simulate the situations with less and less training cells